Storm surge time series de-clustering using correlation analysis

Abstract

The extraction of individual events from continuous time series is a common challenge in many extreme value studies. In the field of environmental science, various methods and algorithms for event identification (de-clustering) have been applied in the past. The distinctive features of extreme events, such as their temporal evolutions, durations, and inter-arrival times, vary significantly from one location to another making it difficult to identify independent events in the series. In this study, we propose a new automated approach to detect independent events from time series, by identifying the standard event duration across locations using event correlations. To account for the inherent variability at a given site, we incorporate the standard deviation of the event duration through a soft-margin approach. We apply the method to 1 485 tide gauge records from across the global coast to gain new insights into the typical durations of independent storm surges along different coastline stretches. The results highlight the effects of both local characteristics at a given tide gauge and seasonality on the derived storm durations. Additionally, we compare the results obtained with other commonly used de-clustering techniques showing that these methods are more sensitive to the chosen threshold.